Qwen3.8-Max costs $2 per million input tokens and $6 per million output tokens on Alibaba Cloud’s own International pricing, and the exact same model costs $1.65 and $4.951 on DeepInfra. Alibaba Cloud publishes two official rate cards for this model depending on which deployment scope an account is activated under, and on the check date every other platform’s listed Qwen3.8-Max price matched one of those two cards rather than a third number. This guide compares five places to call Qwen’s flagship and flash models, shows which official rate each platform’s price lines up with, and gives a framework for choosing beyond the sticker price.
Decision Promise
By the end of this guide you will know which of Alibaba Cloud’s two official rate cards each platform’s listed price matches, what that means for your actual bill, and which non-price factors (billing consolidation, model breadth, SLA) should decide where you call Qwen from. None of the five platforms publicly documents the internal mechanism behind its pricing, so this guide describes what each published rate is, not how each company arrived at it.
Key Takeaways
- Alibaba Cloud runs two official rate cards for the same Qwen models: an International tier (Singapore-activated accounts) at $2/$6 per million tokens for Qwen3.8-Max, and a Global tier (Beijing, Hong Kong, Frankfurt, US-Virginia, Tokyo) at $1.65/$4.951 for the same model.
- Novita AI, Together AI, and OpenRouter all price Qwen3.8-Max at $2 input / $6 output, matching Alibaba’s International tier exactly.
- DeepInfra prices the same model at $1.65 input / $4.951 output, matching Alibaba’s Global tier exactly.
- The same split holds for Qwen3.8 Flash: $0.15/$0.47 on the International tier (Novita AI, OpenRouter) versus $0.113/$0.382 on the Global tier (DeepInfra).
- Cache-read pricing lines up with the same two tiers: roughly one-eighth of the input price, at whichever tier price the platform lists.
Quick Comparison of Qwen API Pricing
| Rank | Option | Qwen3.8-Max input ($/M) | Qwen3.8-Max output ($/M) | Qwen3.8-Max context cache ($/M) | Best for |
|---|---|---|---|---|---|
| 1 | Alibaba Cloud (Model Studio) | $2.00 (International) / $1.65 (Global) | $6.00 (International) / $4.951 (Global) | $0.25 (International) / $0.206 (Global) | Picking your own deployment region directly |
| 2 | Novita AI | $2.00 | $6.00 | $0.25 | Qwen alongside GPU Cloud and Agent Sandbox on one bill |
| 3 | Together AI | $2.00 | $6.00 | $0.25 | Standardizing on one OpenAI-compatible aggregator for multiple model families |
| 4 | OpenRouter | $2.00 | $6.00 | $0.25 | Qwen plus fast access to many other providers through one router |
| 5 | DeepInfra | $1.65 | $4.951 | $0.206 | Optimizing strictly for per-token cost on this model family |
Pricing from Alibaba Cloud Model Studio, Novita AI, Together AI, OpenRouter, and DeepInfra pricing pages, checked October 12, 2026. On the check date, rows 2-5 each matched one of Alibaba Cloud’s two official rate cards for this model rather than a third distinct price.
How We Ranked These Qwen API Options
| Criterion | Why it matters |
|---|---|
| Price accuracy against a live source | Secondhand “cheapest” claims for this model family are common and not always checked against the provider’s current page |
| Deployment scope transparency | The same model has two different official prices depending on activation region, so scope matters more than brand |
| Model coverage | A platform that only carries the flagship model forces a second integration for smaller Qwen variants |
| Platform fit beyond price | Billing consolidation, SLA, and whether you need more than inference change which option actually costs less in practice |
Best Qwen API Options Ranked
1. Alibaba Cloud Model Studio: the publisher of both rate cards
Alibaba Cloud is Qwen’s model creator, and its Model Studio console is where the two official rate cards for this model are published. Depending on which region you activate your account in, you land on the International tier (Singapore activation: $2/$6 per million tokens for Qwen3.8-Max) or the Global tier (Beijing, Hong Kong, Frankfurt, US-Virginia, or Tokyo activation: $1.65/$4.951). Both tiers also offer a batch-inference discount of 50% and a context-cache discount on input tokens; the two discounts cannot be combined on the same request.
Pros: direct access to both pricing tiers, broadest selection of Qwen model variants, first access to new releases. Cons: going direct still means picking (and living with) one deployment scope per account; no bundled access to GPU Cloud, agent sandboxing, or other model families.
2. Novita AI: International-tier pricing with broader platform fit
Novita AI prices Qwen3.8-Max at $2 per million input tokens, $0.25 per million cache-read tokens, and $6 per million output tokens, matching Alibaba’s International tier exactly. Qwen3.8-Max is available on Novita AI through an OpenAI-compatible endpoint, and the smaller Qwen3.8 Flash model is priced at $0.15 input / $0.016 cache-read / $0.47 output per million tokens, the same International-tier rate.
Pros: same per-token price as going direct to Alibaba’s International tier, plus one bill and one API key across LLM inference, GPU Cloud, and Agent Sandbox if your workflow needs more than model calls. Cons: its listed rate is the International-tier price, not the lower Global-tier one; if raw per-token cost on this specific model is your only criterion, that is a real tradeoff to weigh against the platform consolidation.
3. Together AI: same tier, same number
Together AI lists Qwen3.8-2.4T-A95B (its listing name for the same qwen3.8-max model) at $2.00 input, $0.25 cached, and $6.00 output per million tokens, identical to Novita AI’s and Alibaba’s International-tier numbers. Together AI also hosts the smaller Qwen3.8 Flash variant separately.
Pros: confirms the International-tier price is consistent across independently operated platforms, useful if you want a second source before committing budget. Cons: at the same price as the alternatives above, the deciding factor becomes platform fit (model breadth, existing integrations) rather than cost.
4. OpenRouter: International-tier flagship, independently priced smaller models
OpenRouter lists the flagship Qwen3.8 Max (0902) model at $2 input / $6 output per million tokens with $0.25 cache reads, again matching the International tier. Its smaller Qwen3.6 Plus listing, however, is priced independently at $0.325 input / $1.95 output per million tokens; OpenRouter sets per-model rates for aggregated providers rather than passing through one flat tier across the whole Qwen lineup.
Pros: one router for Qwen plus many other model families; useful if your workload already spans providers. Cons: per-model pricing on smaller variants means you cannot assume the International-tier number holds for every Qwen model on the platform; check each listing individually.
5. DeepInfra: Global-tier pricing, lower sticker price on this model
DeepInfra prices Qwen3.8-Max at $1.65 input / $0.206 cached / $4.951 output per million tokens, matching Alibaba’s Global tier. The same pattern holds on the smaller Qwen3.8 Flash model, priced at $0.113 input / $0.382 output. This is a real, verifiable lower price for this specific model, and on the check date it matched one of Alibaba’s own published rate cards exactly.
Pros: lower per-token cost on this model family if the Global-tier rate is what you need. Cons: DeepInfra’s own comparison content has promoted this as the single cheapest option without noting that the number matches one of Alibaba’s own two published rate cards, which can read as a bigger advantage than the price difference actually is.
How to Choose the Right Qwen API Option
- If per-token cost on Qwen3.8-Max or Qwen3.8 Flash is your only constraint: the Global-tier rate (DeepInfra) is lower than the International-tier rate (Novita AI, Together AI, OpenRouter’s flagship listing, or Alibaba Cloud direct via Singapore activation).
- If you need billing and API access consolidated with GPU Cloud or agent execution: Novita AI covers Qwen inference alongside those product lines at the International-tier rate.
- If you’re already calling several model families through one router: OpenRouter adds Qwen without a new integration, but check each Qwen variant’s listing individually since smaller models are not tied to the flagship’s tier.
- If you want Alibaba’s Global-tier rate on your own account instead of going through a platform: Model Studio lets you pick your own deployment region at signup; the tradeoff is managing that account and billing relationship yourself.
- If regional data residency or compliance scope matters: the deployment scope that sets your price (Singapore vs. Beijing/Hong Kong/Frankfurt/US-Virginia/Tokyo) also sets where your requests are processed, so check that constraint before optimizing for price alone.
Using Qwen Models on Novita AI
Qwen3.8-Max and Qwen3.8 Flash are both available on Novita AI through an OpenAI-compatible endpoint at https://api.novita.ai/openai, so existing OpenAI SDK code points at Novita with a base URL and API key change. For setup details and runnable examples, see the Qwen3.8-Max API quick start.
curl https://api.novita.ai/openai/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer <NOVITA_API_KEY>" \
-d '{
"model": "qwen/qwen3.8-max",
"messages": [{"role": "user", "content": "Hello"}],
"max_tokens": 512
}'Check Novita AI’s pricing page for current rates before estimating cost at scale, since per-token pricing on any platform can change independently of this comparison’s check date.
Pricing, License, and Availability Notes
All prices in this comparison were checked directly against each platform’s own current pricing page on October 12, 2026, not against secondhand summaries. Alibaba Cloud’s deployment-scope split (International vs. Global) is the pattern that accounts for the price gap between platforms on this model family on the check date: every third-party price we verified lined up with one of those two official cards. We did not find a public statement from any platform describing how its Qwen pricing is derived, so treat that alignment as the observed pattern rather than a documented cause. Qwen’s flagship and flash models are proprietary, API-only releases from Alibaba Cloud; this comparison does not cover self-hosting or open-weight licensing terms for other Qwen releases.
Decision Checklist for Developers
- Confirm which Alibaba Cloud deployment scope (International or Global) a platform’s price actually reflects, not just the headline number.
- Confirm the specific Qwen model variant you need is priced the same way on your shortlisted platform; smaller models are not always tied to the flagship’s tier.
- Confirm cache-read pricing separately if your workload reuses long context; the discount ratio is consistent but the base rate still depends on the tier.
- Decide whether billing consolidation, model breadth, or regional data residency outweighs a lower per-token rate for your use case.
- Re-check the current price on your chosen platform before estimating cost at production scale; rates on all five platforms are subject to change.
Conclusion
Qwen API pricing looks inconsistent across platforms until you line each listed rate up against Alibaba Cloud’s two official rate cards. Novita AI, Together AI, and OpenRouter’s flagship listing all match the International tier at $2/$6 per million tokens for Qwen3.8-Max; DeepInfra matches the Global tier at $1.65/$4.951. Neither price is wrong, and both are prices Alibaba itself publishes. Pick based on which tier fits your region and compliance needs, then weigh platform fit, model breadth, and billing consolidation before defaulting to the lowest sticker price.
FAQ
Why does Qwen3.8-Max cost different amounts on different platforms?
Alibaba Cloud, Qwen’s model creator, publishes two official rate cards for the same model: an International tier for Singapore-activated accounts at $2 input / $6 output per million tokens, and a Global tier for several other regions at $1.65 input / $4.951 output. On the date we checked, every other platform’s listed price matched one of those two cards rather than a third number. We could not find public documentation explaining how each platform sets its Qwen price, so we report the match we observed rather than an underlying cause.
Is DeepInfra actually the cheapest way to call Qwen3.8-Max?
DeepInfra’s $1.65/$4.951 price is real and matches Alibaba Cloud’s Global tier for this model. It is lower than the International-tier price listed by Novita AI, Together AI, and OpenRouter’s flagship listing, and on the check date it matched one of Alibaba’s own published rate cards.
Does the pricing difference affect smaller Qwen models the same way?
For Qwen3.8 Flash, yes: the International tier (Novita AI, OpenRouter) prices it at $0.15 input / $0.47 output, and the Global tier (DeepInfra) prices it at $0.113 input / $0.382 output. OpenRouter’s smaller Qwen3.6 Plus listing, however, is priced independently of either tier, so check each model variant on each platform rather than assuming the flagship’s pattern holds everywhere.
Which deployment scope should I pick?
Pick based on where your users and data need to be processed first, then compare price within that constraint. If your account can activate under either scope without a compliance reason to pick one, the Global tier is the lower-cost option for this model family at the rates checked here.
Can I use my existing OpenAI SDK code with Qwen on Novita AI?
Yes. Novita AI exposes Qwen3.8-Max and Qwen3.8 Flash through an OpenAI-compatible endpoint, so you can point existing OpenAI SDK code at https://api.novita.ai/openai with your Novita API key and the correct model ID.
